A model based on the gradient boosting regressor to predict trends in the ratio of residence in relation to the age of homeless people in Colombia

Objective: To generate a machine learning model that is capable of predicting trends in the reasons that lead to being a street dweller during adolescent and adult ages. Methodology: The model is trained with information of the Census of Street Dwellers - CHC- 2021 of the National Administrative Department of Statistics (DANE), which contains 19.375 records and 22 variables. The population of street dwellers is identified to include children, young people, adults, the elderly, and even families, who, regardless of their age, gender, race, marital status, social or mental condition, or occupation, live there depending on a reason that somehow forces them to stay permanently or for periods of time. Then, different models were applied until the results described below were obtained. Results: The paper presents an ensemble model of machine learning algorithms based on Gradient Boosting Regressor to predict trends in the reason for street dwelling in relation to the age of street dwellers in Colombia. The results obtained in the model evaluation are promising, providing validity for the model to serve as a basis for government institutions for the formulation, management, and evaluation of policies, plans, and programs of the municipal administration in relation to street dwellers. Conclusions: It can be concluded that the proposed model can serve as a basis to support decision-making by government institutions for public policies, and programs of the municipal and local administration regarding the comprehensive care, rehabilitation, and social inclusion of street dwellers in Colombia.